drizzle orm is overhyped
Everyone is talking about drizzle orm. we all know that vercel pays people to talk
I feel prisma DX is no match for drizzle orm no matter what they say prisma is here to stay
You may say performance
Why not develop you backend in rust or golang if performance was your main focus
DX and community 🎤
38 Replies
🤡
You may say performance
Why not develop you backend in rust or golang if performance was your main focus
like
performance in sql terms
not in raw runtime performance
if you run the same query, be on nodejs, rust, golang or whatever
can you guess the performance?Most applications won't reach the so called performance bottle necks
The term is performance is overshoot so as scalability leading to over engineering
Prisma is enough for most applications but this whole drizzle hype may pressure you people to rebuilding it all
I will say that when it comes to prisma ther performance actually IS an issue at any scale
cause no matter what size your app is, if you run it serverlessly youre gunna eat a huge coldstart time which matters much more if your app is small
prisma kinda seems to reverse scale in that way. it gets BETTER the more users you have cause you get less cold starts
prisma has a great DX, but its also a full ORM which has advantages and disadvantages. The main disadvantage is you do kinda get locked in hard.
for reference, im using kysely for my app which i like a lot because its very easy to understand: it makes my sql queries typesafe. nothing more and nothing less
i dont really use the migrations, but they seem overrated a bit overrated in both drizzle and prisma. for instance in prisma, (AFAICT) i need to specify default values to add a column. i cant do calculations on previous columns.
maybe im wrong about that last bit, by the point i was looking into migrations i was already looking elsewhere because of the coldstart performance
(also the fact that it manually does its own joins is bonkers)
Drizzle has a long way to go, agreed.
Dose it worth the pain? YES!
I was using Prisma for a db with 30 tables. To simply get the context for the user, it created 4 different queries. Each request would take more then a second just to get the context.
With Drizzle using prepared statement, this is down to 60ms.
I'm not hyped, just practical.
as far as i’ve understood, its a great alternative when ur going serverless. so i decided to go with bare metal and go with prisma…
Not having to generate types and restart your lsp when changing your schema is good too haha
Yes, cold starts and edge are the main selling points, but Drizzle will speed up your queries in any env - especially with db like PlanetScale.
I see.
prisma is over engineered 
Eventually your typescript server will die and you will still restart it😅
I would say prisma's engineers.... Would not say it's engineered with how poor it actually does queries
not having to run codegen in my build is a big +
Agree.
Please mind you I am saying what I said above .... But I am using Prisma
i thoguht recently prisma made some changes that made them like 10x faster or something for coldstart on serverless.
is Drizzle still miles better than PRisma?
They work differently
Prisma still have giant query engine that everything goes trough
And not best thing in optimized queries so more complex queries will still hurt a lot
i tried these changes (which were not well documented btw). they did not make it 10x faster. i couldnt replicate the results they listed in the blog post
and the fact that the orm does manual joins is bonkers
i get that it was probably easier to develop, but that should be a minimum requirement for an orm
better is also relative. Drizzle is still in active development so its not gunna have all the features and stability
but it definitely performs way better
and you may not need the features
Why is this in questions lol
@barry this is the first time you can tell someone to not post in questions
@Lopen32
You done fucked up little lopen man, this is the questions channel, for questions, we have a #tech-discussion for this kind of bullshit
Interesting!
I'm assuming you're using drizzle in a vercel serverless environment.
how much of a performance increase do you see/feel for basic operations?
im actually using kysely now. i tried drizzle but opted for kysely for some more structural reasons
but my test db has literally 1 row in it and i just did a query on that row and on cold starts i saw it take up to 9s to return data
kysely cold start is still in the 4s range but thats at least tolerable for something i dont plan on happening that often once the app scales up a bit
i think almost all of that cold start time is actually vercel / aws though
1 very frustrating thing is that it seems near impossible to get good accoounting of where all the time is going
so i cant really say for sure why kysely is so much faster, but i can point to the numbers and make some educated guesses
?
Just emphasizing 🙂
emphasizing what? im sorry i dont understand
he's agreeing width you and showing an example of a query with kysely ( that took 4ms )
showing how fast it is
Oh. I didn’t see that in the picture
Thanks
Sorry was on a call, yeah what @Diogovski said.
Its taken from twitter, when it was first falsly claimed that only drizzle hits single digit latency, and later corrected (issue was misconfiguration on deployment of non-drizzle examples)
most of the latency is not even kysely related, but communication/compute between lambda/worker and db
yeah i mean kysely is just a very nice string builder really so that makes sense
side note. it would be really nice if there was a way to examine the complete timeline of a lambda
even when i was using prisma it would report something like 20ms for the query, but the whole request still took 9 seconds on lambda
The term is performance is overshoot so as scalability leading to over engineering
P R E A C HThe thing is i use a real server for my deployments
If you don't use serverless then use what you like??? I have two apps that utilize prisma that are hosted on Render.com; great performance. Yet it does suck incredibly on serverless apps, like, bad.
Each has their pros and cons; but there's no arguing that if severless continues to increase in popularity then either prisma needs to figure out their cold starts or it will lose popularity pretty quickly. I'd love to use prisma on serverless apps but I cannot risk churn through 6 second cold starts vs Drizzle's 100ms
I thought Serverless infrastructure was the cause of the cold starts, but here the picture your painting makes it seems like the orm is the source of the cold start
Partially true, dont get me wrong I personally prefer prisma too, but prisma does add a huge over head on top of the coldstarts when you ur using serverless...
drizzle is a step back in humanity. Prisma lets u do many to many connections with ease while in drizzle u still have to do the old school way of creating a middle table like a pleb.
And the columns are not even lined up in a nice way like in prisma's schema. The way forward for ORMs are custom languages for schemas like prisma does. everything else is poop
prisma also gives u convinient item arrays that dont exist behined the scenes
Use #tech-discussion for discussions & debates (be sure to keep them civil and logical)